[PATCH libX11 1/2] xcb_io: Fix Xlib 32-bit request number wrapping

[Jonas Petersen]

Am 18.11.2013 19:16, schrieb Julien Cristau:
> On Sun, Nov 17, 2013 at 21:30:44 +0100, Jonas Petersen wrote:
>
>> Am 17.11.2013 20:20, schrieb Mouse:
>>>> I guess the sizeof comparison would not be necessary since the
>>>> condition should never meet with 64-bit longs.
>>> Unless it's in a code fragment that's used only on machines with
>>> <64-bit longs, it will; X runs on systems with 64-bit longs.
>> I meant the condition "dpy->request < dpy->xcb_last_flushed" should
>> not meet on systems with 64-bit longs (at least not until the 64-bit
>> wrap). So the "sizeof(uint64_t) > sizeof(unsigned long)" would not
>> be necessary. It would just increase the overhead on <64-bit
>> systems.
>>
> The sizeof comparison is a compile-time constant so I'd expect it to be
> optimized out by the compiler anyway, so no overhead.

This makes sense. Being curious, I did some tests (with gcc 4.7.2 on 
i686). They showed that this code:

     int i = 0;
     for (;;) {
         ++i;
         if (sizeof(unsigned long) > sizeof(unsigned int) && i < 100) {
             printf("%d\n", i);
         }
     }


on a 32-bit system compiles into the same binary as this code:

     for (;;) {
     }

and on a 64-bit system it compiles into the same binary as this code:

     int i = 0;
     for (;;) {
         ++i;
         if (i < 100) {
             printf("%d\n", i);
         }
     }

All that without any optimization options.

So adding the size comparison seems to not only not increasing overhead, 
but even effectively optimizing for the targets. At least with this 
compiler on these architectures.

- Jonas